LETTERS

PUBLISHED ONLINE: 10 JUNE 2012 | DOI: 10.1038/NNANO.2012.83

Self-orienting nanocubes for the assembly of plasmonic nanojunctions

Bo Gao, Gaurav Arya and Andrea R. Tao*
Plasmonic hot spots are formed when metal surfaces with high curvature are separated by nanoscale gaps and an electromagnetic eld is localized within the gaps. These hot spots are responsible for phenomena such as subwavelength focusing1,2, surface-enhanced Raman spectroscopy3 and electromagnetic transparency4, and depend on the geometry of the nanojunctions between the metal surfaces5. Direct-write techniques such as electron-beam lithography can create complex nanostructures with impressive spatial control6 but struggle to fabricate gaps on the order of a few nanometres or manufacture arrays of nanojunctions in a scalable manner. Self-assembly methods, in contrast, can be carried out on a massively parallel scale using metal nanoparticle building blocks of specic shape7,8. Here, we show that polymer-grafted metal nanocubes can be self-assembled into arrays of one-dimensional strings that have well-dened interparticle orientations and tunable electromagnetic properties. The nanocubes are assembled within a polymer thin lm and we observe unique superstructures derived from edgeedge or faceface interactions between the nanocubes. The assembly process is strongly dependent on parameters such as polymer chain length, rigidity or grafting density, and can be predicted by free energy calculations. A signicant challenge in the self-assembly of shaped nanoparticles is the formation of non-close-packed nanoparticle groupings that adopt specic interparticle orientations, which precludes the use of assembly methods that involve colloidal crystallization or jamming. Several strategies have been developed for achieving programmable assembly through anisotropic chemical modication of the nanoparticle surface, including the use of DNA linkers9,10, block co-polymers11,12 or patchy particles1315. However, these approaches are solution-based, require extensive chemical modication of nanoparticle surfaces, and ultimately necessitate post-processing to transfer the nanoparticle assemblies into a dielectric material or onto a solid support. Our approach to fabricating large-area, non-close-packed plasmonic nanoparticle arrays is to organize polymer-grafted metal nanoparticles directly within a supported polymer thin-lm matrix. This multicomponent system of nanoparticles, grafted polymers and matrix polymer is expected to yield rich phase behaviour due to non-specic nanoparticle interactions arising from van der Waals and steric forces. Self-assembly can be tuned using simple parameters such as polymer chain length, graft density and miscibility. Previously, graft chain length has been demonstrated as a key parameter in dictating spherical nanoparticle assembly into anisotropic superstructures ranging from aggregates to sheets and strings16,17. Here, we demonstrate how graft chain length can be used to control the orientation of shaped nanoparticles within hierarchical nanojunctions. This simple yet powerful approach does not require any chemical anisotropy of the nanoparticle surface, thus simplifying surface modication and forgoing the need for faceselective surface chemistries18. Because assembly is not guided by strong chemical bonds between nanoparticles, our strategy allows for nanoparticle assemblies that are dynamic and can be engineered to exhibit stimulus-responsive electromagnetic behaviour. Moreover, the polymer matrix serves as a convenient dielectric medium that can be used to freeze-capture desired nanoparticle structures as they evolve. We fabricated a spin-on plasmonic lm using nanoparticles organized within large-area (.6 cm2) polystyrene thin lms. The basic building block for our plasmonic layer is a metal nanocube grafted with hydrophilic polymer chains. Silver nanocubes were batch synthesized using wet chemical methods to produce colloidal dispersions of uniform particle size and shape19,20 (see Supplementary Section S1 for full details). The cubic shape is unique in two ways. First, it is a compelling geometry for constructing non-close-packed nanoparticle architectures by coordination through facet, corner or edge sites. In the thin lms fabricated here, this anisotropy leads to assembly primarily through edgeedge, faceface and faceedge orientations (Fig. 1a). Second, this shape supports the excitation of higher-order surface plasmon modes that occur through charge localization into the corners and edges of the nanoparticle. This enables orientation-dependent electromagnetic coupling between neighbouring nanoparticles21, where interparticle junctions formed by cube corners and edges produce intense electromagnetic elds that are conned well below the conventional diffraction limit. To induce spontaneous organization, the hydrophilic polymergrafted nanocubes are introduced into a hydrophobic polymer matrix. In the spinodal regime22, this immiscible blend undergoes spontaneous phase segregation to produce periodically spaced and highly aligned one-dimensional strings of nanoparticles. To maximize electromagnetic coupling within these string structures, we rst targeted an edgeedge conguration by grafting the silver nanocubes with a long, oppy polymer ligand. Figure 1b shows this process for nanocubes (edge length, 80 nm) grafted with poly(vinyl pyrrolidone) (PVP, Mw 55,000) and embedded within a polystyrene (Mw 10,900) thin lm with a thickness of 150 nm. As the lm is annealed using thermal or solvent vapour treatment, the nanoparticles assemble in the edgeedge conguration to form strings that continuously grow and converge (Fig. 1ce). The formation of strings is likely to result from two effects: large steric repulsion due to an effective increase in graft density locally at nanojunction sites23, and diffusion-limited assembly resulting from low nanoparticle mobility within the viscous polymer matrix24. However, these mechanisms for long-range order operate independently of the short-range interparticle interactions that give rise to specic nanocube orientations within an assembled nanojunction. To provide insights into the experimental parameters that dictate the orientation of our nanocubes, we performed Monte Carlo simulations of polymer-grafted nanocubes treated at a coarse-grained resolution (see Supplementary Section S3 for full details of the computational methods). Our approach computes the potential of mean

NanoEngineering Department, University of California, San Diego, La Jolla, California 92093-0448, USA. * e-mail: atao@ucsd.edu
NATURE NANOTECHNOLOGY | VOL 7 | JULY 2012 | www.nature.com/naturenanotechnology

433

2012 Macmillan Publishers Limited. All rights reserved.

LETTERS
a
S Ag S NCS S S SAg SNC S Ag NC S S S S S S S S S S Ag NC

NATURE NANOTECHNOLOGY
a
d

DOI: 10.1038/NNANO.2012.83

b
5 0 5 10 5 0 5 10 5 0 5 10 5 0 5 10 16 80

0 20 80 90

L=4

Potential of mean force (1018J)

L=8 FF EE

c
Edgeedge nanojunction Faceface nanojunction
PS

L = 12

0.8 FF 0.6 0.4 EE

b
150 nm Polystyrene Ag NC embedding

= PVP

L = 16

NCpolymer segregation

0.2 4 8 L 12

85

90 d (nm)

115

120

Figure 1 | Self-oriented nanocrystals for plasmonic materials. a, Schematic of assembly of silver nanocubes (NCs) into edgeedge and faceface orientations by modifying the surface of the nanocubes with polymer grafts of different chain lengths. b, Schematic of the assembly process for edge edge nanojunctions. During the solvent annealing process, PVP-grafted nanocubes are embedded into a polystyrene thin lm. The nanoparticle polymer composite then undergoes spontaneous phase segregation to form nanocube strings. c, Nanocube strings assemble uniformly across the entire lm as observed by dark-eld optical microscopy. The texture observed in this image results from individual nanocube strings (green areas) distributed within the polymer matrix (black areas), but resolution is limited by the diffraction limit. Variations in colour are derived from strings with different lengths and branching structures. Scale bar, 10 mm. d,e, SEM images of the edgeedge-oriented nanocubes within the polystyrene matrix. Scale bars, 1 mm (d) and 200 nm (e).

Figure 2 | Theoretically predicted interaction between silver nanocubes. a, Representative snapshots of faceface- and edgeedge-oriented nanocubes grafted with polymer chains of length L 8 beads. Chains on the two nanocubes are coloured differently for clarity. b, Computed potential of mean forces W for faceface (FF, blue lines) and edgeedge (EE, red lines) oriented nanocubes grafted with varying chain lengths L and xed polymer surface interaction strength (1PS 0.4) plotted as a function of the centreto-centre separation distance d. The inset for L 4 depicts the entire energy well that occurs for the faceface conguration, which is signicantly deeper than for the other calculated chain lengths. c, Phase diagram showing the most favourable nanocube orientation for varying L and 1PS , as predicted from our PMF calculations. The dashed line symbolizes the boundary between the faceface and edgeedge phases.

force (PMF) between two nanocubes oriented in the faceface and edgeedge congurations as a function of their centre-to-centre separation distance d (Fig. 2a). The PMF, which gives the free energy of interaction between nanocubes, allows us to determine the most favourable orientation and equilibrium separation distance between two cubes. We examined the effects of varying the length L of the grafted polymer chains and the effective strength of attraction 1PS with the nanocube surface. Other system parameters did not produce an appreciable effect on the interactions between the nanocubes (such as grafted polymer interaction, 1PP) or are difcult to tune experimentally (such as matrix polymer interactions). Figure 2b shows the computed PMF of two nanocubes in face face (red line) and edgeedge (blue line) orientations grafted with polymer chains of varying length L as a function of separation distance d. When the grafted chains are short (L 4 monomer units), it is energetically favourable for the nanocubes to orient in the face face conguration. The faceface-oriented nanocubes exhibit a strong van der Waals attraction that exceeds the steric repulsion between chains, yielding a deep energy minimum (Supplementary Fig. S4a). In contrast, the edgeedge-oriented nanocubes exhibit a weak van der Waals attraction and negligible steric repulsion,
434

yielding a shallower energy minimum compared with the face face arrangement (Supplementary Fig. S4b). When the polymers grafted to the nanocubes are long, the edgeedge conguration now becomes more energetically favourable. This is because the steric repulsion between these longer polymer chains exceeds the van der Waals attraction in the faceface-oriented nanocubes, making the net free energy repulsive (Supplementary Fig. S4c). In contrast, the edgeedge-oriented nanocubes still exhibit weak steric repulsion and van der Waals attraction, leading to a moderate but net attractive free energy prole (Supplementary Fig. S4d). The interactions between grafted chains and the nanocube surface also play an important role in establishing the most favourable nanocube orientation. When the effective energy parameter between the polymer chains and the metal surface (1PS) is small, the edgeedge conguration becomes more energetically favourable with increasing polymer chain length compared with the faceface conguration. When 1PS is strong, this transition occurs at longer L (Supplementary Fig. S5). This result can be intuitively understood by considering that the faceface conguration allows for signicantly more favourable polymersurface contacts than the edge edge conguration. This interplay between chain length and polymersurface interactions gives rise to the assembly phase diagram of Fig. 2c, which summarizes the favoured orientation of the nanocubes over a broad range of L and 1PS. To demonstrate that this phase diagram can be used to tailor nanoparticle orientation, we induced the rearrangement of polymer-embedded cubes from the edgeedge to the faceface conguration. We chose to focus on tuning polymer chain length L, which can be explicitly dictated by the choice of nanoparticle ligand and where ligand binding to metal nanoparticles can be accomplished using simple thiol chemistry. The faceface

NATURE NANOTECHNOLOGY | VOL 7 | JULY 2012 | www.nature.com/naturenanotechnology

2012 Macmillan Publishers Limited. All rights reserved.

NATURE NANOTECHNOLOGY
a
= PEG-thiol Solvent annealing EE

DOI: 10.1038/NNANO.2012.83

LETTERS
d
100
S S

Thermal annealing

FF

Ag Ag

80

PVP

Population (%)

PEG

60 40 20 0

Face Edge

Solvent

Solvent + thermal PVP

Solvent

Solvent + thermal PEG

Figure 3 | Chain-length-dependent orientation of nanocube junctions. a, Schematic of the assembly of PEG-grafted nanocubes. On thermal annealing, the nanocubes rearrange from the edgeedge (EE) to faceface (FF) conguration. b,c, SEM images of nanocubes modied with PVP-thiol after solvent annealing (b) and PEG-alkanethiol ligands after both solvent annealing and thermal annealing at T . Tg for 4 h (c). On thermal treatment, the PEG-modied nanocubes are converted to assemblies formed through face coordination. Scale bars (b,c), 1 mm; inset scale bars (b,c,), 100 nm. d, Per cent of edge and face coordination in nanocube assemblies before and after thermal treatment. Nanocubes modied with PVP experience little to no change in orientation, whereas nanocubes modied with PEG experience almost complete orientation rearrangement.

orientation requires nanoparticle grafts that are rigid and have relatively short chains. We chose to modify the silver surface with (11-mercaptoundecyl)tetra(ethylene glycol), an alkanethiol terminated with a four-monomer polyethylene glycol (PEG) chain. (See Supplementary Section S4 for full details of surface modication.) These grafts are expected to form a densely packed molecular monolayer, but still allow interdigitation between the matrix polymer and the PEG moieties. As a control, we tested nanocubes chemically modied with thiol-terminated PVP (Mw 1,700). Based on monomer unit persistence lengths, we estimate that the PEG chains have an end-to-end distance of 2.0 nm (ref. 25), in comparison to PVP chains, which exhibit a length of 5.7 nm. Our Monte Carlo simulations predict that the PEG-modied cubes will adopt the edgeedge conguration at large interparticle distances and will reorient to adopt the faceface conguration at a shorter, critical distance (Fig. 2b, top plots). To prevent cubes from becoming kinetically trapped in edgeedge assemblieswhich represents a local free energy minimum but not the global minimum lms were subjected to a two-step annealing process, rst by exposure to chloroform vapour, resulting in the formation of edgeedge assemblies, and second by thermal treatment above Tg of the polystyrene matrix to rearrange these assemblies into the faceface conguration (Fig. 3a). Figure 3b,c presents scanning electron microscope (SEM) images of plasmonic lms containing PVP- and PEG-grafted nanocubes, respectively. Before thermal treatment, 15.1+1.9% of PEG-grafted cubes are in the faceface orientation and 68.6+2.2% are in either the edgeedge or face edge orientations (Supplementary Fig. S8). After thermal treatment, the particles remain assembled in one-dimensional domains but with 88.7+1.8% of the particles oriented in the faceface conguration (Fig. 3d). When grafted with the short PEG chains, thermal activation is able to overcome the energy barrier for cube rotation from the edgeedge to the faceface conguration. As conrmation, faceface assembly is not observed without PEG modication or without thermal activation of the nanocubepolymer lm. Additional assembly experiments using long (Mw 5,000) and extra-long (Mw 54,000) PEG grafts were carried out to conrm the effect of chain length and to rule out the contribution of specic chemical interactions in this reorientation. The use of extra-long

PEG grafts resulted in nanocube assemblies with a large percentage of edgeedge-oriented nanojunctions. However, we observed defects attributed to partial dewetting of the matrix polymer near the cube surface on solvent vapour exposure. Extra-long PEG grafts tend to collapse into globular structures, which may contribute to the poor interaction between the graft and the matrix polymer. This was ameliorated by using long PEG grafts with a lower molecular weight and by thermal annealing to promote polymer redistribution. Nanocubes modied with these longchain PEG grafts assemble predominantly in the edgeedge conguration and remain in the edgeedge conguration after extended periods of thermal treatment (see Supplementary Section S7 for the details of these results). The surface plasmon modes supported by nanoparticlepolymer lms in these two nanojunction orientations are expected to produce different eld localization effects, as predicted by electrodynamic simulations for nanocube dimers (Fig. 4a). This orientation dependence may provide a convenient strategy for tuning the optical properties of the large-scale lms (Fig. 4b) or the surface plasmon coupling properties of individual nanoparticle strings. To probe this experimentally, we measured the optical extinction spectra for lms containing PVP-grafted nanocubes oriented in the edgeedge conguration and PEG-grafted nanocubes in the faceface conguration. The optical response of each lm changes dynamically with increased annealing time as the nanoparticles assemble and form longer string structures (Fig. 4c,d). String length was tuned by varying the solvent exposure times for each sample, and a detailed analysis of string length is given in Supplementary Table S2. Optical spectra were collected by measuring the transmission of unpolarized light over a sample area of 0.8 cm2 to demonstrate the uniformity of nanoparticle orientation across a macroscopic area. Compared with the plasmon response generated by isolated nanocubes before assembly (Supplementary Fig. S9), both lms exhibit scattering peaks at l . 600 nm attributed to strong electromagnetic coupling and the generation of new coupled surface plasmon modes from assembled nanoparticles. Although we observe that these peaks are quite broad (which we attribute to non-uniform string branching), these modes become more intense as the strings grow in length. In the edgeedge conguration, we observe that
435

NATURE NANOTECHNOLOGY | VOL 7 | JULY 2012 | www.nature.com/naturenanotechnology

2012 Macmillan Publishers Limited. All rights reserved.

LETTERS
a
64 32 16 8 4 2 1 100 EE 0 FF Electric eld strength FE 300 200 EE FF 400 Before

NATURE NANOTECHNOLOGY
b

DOI: 10.1038/NNANO.2012.83

c
Optical extinction
EE

y x
60 min 45 min 30 min 15 min

d
Optical extinction
FF

y x
60 min 45 min 30 min 15 min

400

600 800 Wavelength (nm)

1,000

400

600 800 Wavelength (nm)

1,000

ahead, further tuning of 1PS should be able to provide control over string branching and curvature, with the potential to form plasmon-based interconnects or waveguide structures with spatial complexities that are not achievable through top-down approaches. We demonstrate the self-assembly of shaped nanoparticles into non-close-packed arrangements with engineered interparticle junctions. The scalability and ease of fabrication of our approach is valuable to research aimed at the discovery and application of novel plasmonic superstructures. The current limitation of our polymermediated assembly is achieving exact nanoparticle groupings. Our nanocube strings exhibit a distribution of string lengths, in part due to branching. It will be necessary to obtain better kinetic control over the segregation process to achieve precise nanoparticle groupings that can rival current lithographically fabricated patterns. This may be achieved through nanoparticle pinning or by making use of multipolymer blends or block co-polymer matrices to conne microphase segregation to smaller domains. Complementary polymer patterning techniques26 may also be used to direct nanoparticle assembly into superstructures that exhibit more complex curvature, such as split-rings27 or threedimensional assemblies28. These techniques have the potential for generating spin-on optical nanocircuits29, fully addressable plasmonic elements or metamaterial coatings30 and for achieving more complex nanoarchitectures.
Received 7 October 2011; accepted 27 April 2012; published online 10 June 2012

Figure 4 | Plasmonic response of the oriented nanojunctions. a, Electric eld strength (colour scale) for silver nanocubes with edge lengths of 80 nm and a separation gap of 2 nm in three favoured orientations: faceface (FF), faceedge (FE) and edgeedge (EE). Field strengths are plotted in units of V m21 for an incident eld of Eo 1 V m21. Cross-sections are taken in the xy plane bisecting the nanocubes. Field distributions are shown for excitation wavelengths of 750 nm (faceface), 670 nm (faceedge) and 670 nm (edgeedge) corresponding to the coupled surface plasmon mode for each of the nanocube dimers. For visual contrast, the electric eld strength is plotted on a log2 scale with min/max cutoffs as shown in the colour legend. Magnied regions of the nanojunctions are shown in the images on the right, which correspond to an area of 20 nm 20 nm. For visual contrast, the electric eld strength is plotted with min/max cutoffs as shown in the colour legend. The edgeedge conguration supports the strongest eld localization in the narrow interparticle gap, with a maximum eld enhancement of .800Eo in the gap centre and .1,200Eo at the nanocube corners. b, Digital camera images of lms covering a quartz substrate of area 5 cm2 with embedded but unassembled nanocubes (Before), edgeedge-oriented nanocubes (EE) and faceface-oriented nanocubes (FF). c,d, Measured extinction spectra of nanocomposite lms containing PVP-grafted nanocubes in the edgeedge orientation (c) and PEG-grafted nanocubes in the faceface orientation (d). Films in both orientations show peaks corresponding to coupled surface plasmon modes in the near-infrared. Spectra were taken for samples subjected to different solvent vapour exposure times to obtain lms with varying nanoparticle string lengths: 15 min (one particle), 30 min (two particles), 45 min (six particles) and 60 min (15 particles). For the spectra in d, lms were subjected to an additional thermal annealing step to switch the edge edge-oriented strings to faceface-oriented strings. Detailed analysis of string length and representative SEM images are provided in the Supplementary Information.

References
1. Barnes, W. L., Dereux, A. & Ebbesen, T. W. Surface plasmon subwavelength optics. Nature 424, 824830 (2003). 2. Ghosh, S. K. & Pal, T. Interparticle coupling effect on the surface plasmon resonance of gold nanoparticles: from theory to applications. Chem. Rev. 107, 47974862 (2007). 3. Hyunhyub, K., Srikanth, S. & Vladimir, V. T. Nanostructured surfaces and assemblies as SERS media. Small 4, 15761599 (2008). 4. Fan, J. A. et al. Self-assembled plasmonic nanoparticle clusters. Science 328, 11351138 (2010). 5. Stebe, K. J., Lewandowski, E. & Ghosh, M. Oriented assembly of metamaterials. Science 325, 159160 (2009). 6. Gramotnev, D. K. & Bozhevolnyi, S. I. Plasmonics beyond the diffraction limit. Nature Photon. 4, 8391 (2010). 7. Nie, Z., Petukhova, A. & Kumacheva, E. Properties and emerging applications of self-assembled structures made from inorganic nanoparticles. Nature Nanotech. 5, 1525 (2009). 8. Whitesides, G. M. & Grzybowski, B. Self-assembly at all scales. Science 295, 24182421 (2002). 9. Xu, X., Rosi, N. L., Wang, Y., Huo, F. & Mirkin, C. A. Asymmetric functionalization of gold nanoparticles with oligonucleotides. J. Am. Chem. Soc. 128, 92869287 (2006). 10. Millstone, J. E. et al. DNAgold triangular nanoprism conjugates. Small 4, 21762180 (2008). 11. Lin, Y. et al. Self-directed self-assembly of nanoparticle/copolymer mixtures. Nature 434, 5559 (2005). 12. Shenhar, R., Norsten, T. B. & Rotello, V. M. Polymer-mediated nanoparticle assembly: structural control and applications. Adv. Mater. 17, 657669 (2005). 13. Caswell, K. K., Wilson, J. N., Bunz, U. H. F. & Murphy, C. J. Preferential end-toend assembly of gold nanorods by biotinstreptavidin connectors. J. Am. Chem. Soc. 125, 1391413915 (2003). 14. DeVries, G. A. et al. Divalent metal nanoparticles. Science 315, 358361 (2007). 15. Nie, Z. et al. Self-assembly of metalpolymer analogues of amphiphilic triblock copolymers. Nature Mater. 6, 609614 (2007). 16. Akcora, P. et al. Anisotropic self-assembly of spherical polymer-grafted nanoparticles. Nature Mater. 8, 354359 (2009). 17. Maillard, D., Kumar, S. K., Rungta, A., Benicewicz, B. C. & Prudhomme, R. E. Polymer-grafted-nanoparticle surfactants. Nano Lett. 11, 45694573 (2011). 18. Glotzer, S. C. & Solomon, M. J. Anisotropy of building blocks and their assembly into complex structures. Nature Mater. 6, 557562 (2007). 19. Sun, Y. & Xia, Y. Shape-controlled synthesis of gold and silver nanoparticles. Science 298, 21762179 (2002). 20. Tao, A., Sinsermsuksakul, P. & Yang, P. Polyhedral silver nanocrystals with distinct scattering signatures. Angew. Chem. Int. Ed. 45, 45974601 (2006). 21. Lee, S. Y. et al. Dispersion in the SERS enhancement with silver nanocube dimers. ACS Nano 4, 57635772 (2010).

plasmonic coupling results in a long-wavelength surface plasmon that redshifts (from 776 nm to 895 nm) as string length increases. Our simulations for edgeedge-oriented strings indicate that maximum eld localization occurs in the interstitial junction between nanocube edges (Supplementary Fig. S10). In contrast, nanocubes assembled in the faceface conguration behave increasingly like broadband scatterers as string length is increased. This behaviour probably results from surface plasmon delocalization along the extended nanoparticle string, as indicated by simulations showing regions of maximum eld localization smeared over the boundaries of the string. Looking
436

NATURE NANOTECHNOLOGY | VOL 7 | JULY 2012 | www.nature.com/naturenanotechnology

2012 Macmillan Publishers Limited. All rights reserved.

NATURE NANOTECHNOLOGY

DOI: 10.1038/NNANO.2012.83

LETTERS

22. Bates, F. S. Polymerpolymer phase behavior. Science 251, 898905 (1991). 23. Pryamtisyn, V., Ganesan, V., Panagiotopoulos, A. Z., Liu, H. & Kumar, S. K. Modeling the anisotropic self-assembly of spherical polymer-grafted nanoparticles. J. Chem. Phys. 131, 221102 (2009). 24. Li, L. et al. Kinetically trapped co-continuous polymer morphologies through intraphase gelation of nanoparticles. Nano Lett. 11, 19972003 (2011). 25. Zhu, B., Eurell, T., Gunawan, R. & Leckband, D. Chain-length dependence of the protein and cell resistance of oligo(ethylene glycol)-terminated self-assembled monolayers on gold. J. Biomed. Mater. Res. 56, 406416 (2001). 26. Nie, Z. & Kumacheva, E. Patterning surfaces with functional polymers. Nature Mater. 7, 277290 (2008). 27. Kante, B., de Lustrac, A., Lourtioz, J-M. & Gadot, F. Engineering resonances in infrared metamaterials. Opt. Express 16, 67746784 (2008). 28. Agarwal, U. & Escobedo, F. A. Mesophase behaviour of polyhedral particles. Nature Mater. 10, 230235 (2011). 29. Engheta, N. Circuits with light at nanoscales: optical nanocircuits inspired by metamaterials. Science 317, 16981702 (2007).

30. Xiao, S., Chettiar, U. K., Kildishev, A. V., Drachev, V. P. & Shalaev, V. M. Yellow-light negative-index metamaterials. Opt. Lett. 34, 34783480 (2009).

Acknowledgements
A.R.T. acknowledges nancial support from UCSD start-up funds, a NSF BRIGE grant (ECCS-1125789) and a Hellman Fellowship.

Author contributions
B.G. and A.R.T. conceived and designed the assembly experiments. B.G. performed the assembly experiments. G.A. designed and performed Monte Carlo calculations. A.R.T. performed nite element method simulations. G.A. and A.R.T. co-wrote the paper.

Additional information
The authors declare no competing nancial interests. Supplementary information accompanies this paper at www.nature.com/naturenanotechnology. Reprints and permission information is available online at http://www.nature.com/reprints. Correspondence and requests for materials should be addressed to A.R.T.

NATURE NANOTECHNOLOGY | VOL 7 | JULY 2012 | www.nature.com/naturenanotechnology

437

2012 Macmillan Publishers Limited. All rights reserved.